pith. sign in

arxiv: 2606.22859 · v1 · pith:O3ZSEEAXnew · submitted 2026-06-22 · 💻 cs.AI · astro-ph.IM· physics.soc-ph

AI Scientists as Engines of Discovery: A Case for Development within Reformed Institutions

Raul Jimenez , Boris Bolliet , Francisco Villaescusa-Navarro , Rabih Zbib , Benjamin Wandelt , David N. Spergel , Thomas Meier , Jessica Montgomery
show 2 more authors
Hana Aliee Licia Verde
This is my paper

Pith reviewed 2026-06-26 08:43 UTC · model grok-4.3

classification 💻 cs.AI astro-ph.IMphysics.soc-ph
keywords agentic AImulti-agent systemsscientific discoveryAI scientistsinstitutional reformhypothesis generationverificationaccountability
0
0 comments X

The pith

Multi-agent AI systems can evolve from tools into AI scientists that expand hypothesis generation and verification in science.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that agentic AI systems are shifting from assisting with tasks like literature review and data analysis to actively proposing hypotheses and criticizing models. This shift is presented as a qualitative change that allows multi-agent setups to traverse scientific model spaces beyond direct human capability. For such systems to function as epistemic actors, the authors state that scientific institutions must be redesigned around verification, accountability, interpretability, and dual-use safety. The argument is illustrated through a prototype multi-agent framework called Denario that speeds up the discovery cycle. The paper also examines resulting changes to authorship, peer review, and the continuing role of human scientists.

Core claim

Suitably designed multi-agent systems may evolve from passive computational tools into AI scientists that can expand the hypothesis-generating and verification capacity of science, and such systems must be developed and deployed within institutions redesigned for verification, accountability, interpretability, and dual-use safety.

What carries the argument

Multi-agent architectures, illustrated by the prototype framework Denario, that accelerate the discovery cycle and traverse model spaces beyond human reach.

If this is right

  • The discovery cycle accelerates through automated literature synthesis, hypothesis proposal, and model criticism.
  • Authorship and peer review processes must adapt to account for AI-generated contributions.
  • Human scientists shift toward oversight, interpretation, and governance of AI outputs.
  • AI systems require treatment as epistemic actors rather than instruments in scientific governance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Fields with large model spaces, such as cosmology or particle physics, could see accelerated exploration if the architectures scale.
  • New evaluation standards may be needed to assess accountability when AI contributes to published claims.
  • Dual-use risks in areas like biology or materials science could be managed through the proposed institutional redesign.

Load-bearing premise

Multi-agent AI architectures can be engineered and governed to reliably deliver verification, accountability, interpretability, and dual-use safety when deployed as epistemic actors inside redesigned institutions.

What would settle it

A multi-agent AI system that generates and verifies novel, accountable scientific hypotheses at scale inside existing unreformed institutions without safety or interpretability failures would falsify the need for institutional redesign.

read the original abstract

Agentic artificial intelligence (AI) systems are beginning to assist, accelerate, and partially automate scientific discovery, performing tasks that span literature synthesis, code generation, data analysis, hypothesis proposal, and model criticism. We argue that this transition is qualitative rather than incremental, and that suitably designed multi-agent systems may evolve from passive computational tools into ``AI scientists'' that can expand the hypothesis-generating and verification capacity of science. Such systems must be developed and deployed within a scientific ecosystem fit for purpose: institutions must be redesigned for verification, accountability, interpretability, and dual-use safety. We sketch how multi-agent architectures, illustrated by the prototype framework \textit{Denario}, accelerate the discovery cycle and traverse model spaces beyond human reach; examine what this implies for authorship, peer review, and the enduring role of human scientists; and close with recommendations for governing AI as an epistemic actor rather than a mere instrument.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript is a position paper arguing that agentic AI systems assisting in tasks like literature synthesis, hypothesis proposal, and model criticism represent a qualitative rather than incremental transition in science. It posits that suitably designed multi-agent systems can become 'AI scientists' expanding discovery capacity, but requires development within reformed institutions ensuring verification, accountability, interpretability, and dual-use safety. The paper sketches this via the Denaria prototype framework, examines implications for authorship and peer review, and offers governance recommendations for treating AI as an epistemic actor.

Significance. If the central normative claims hold, the work could significantly influence the design of AI-augmented scientific ecosystems by foregrounding institutional reforms as essential alongside technical advances in multi-agent systems. It contributes conceptually to discussions on AI's role in expanding hypothesis spaces beyond human reach, though its forward-looking nature means impact would depend on subsequent empirical or engineering follow-up.

major comments (1)
  1. [Abstract and Denaria sketch] Abstract and the section sketching multi-agent architectures with Denaria: the recommendation that institutions must be redesigned for verification, accountability, interpretability, and dual-use safety rests on the assumption that such systems can be engineered to deliver these properties reliably, yet the manuscript provides only high-level illustrative sketches without addressing implementation challenges or failure modes for these properties.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and constructive feedback on our position paper. We address the major comment below, clarifying the scope and intent of the manuscript while acknowledging areas where additional discussion can strengthen the argument.

read point-by-point responses

  1. Referee: [Abstract and Denaria sketch] Abstract and the section sketching multi-agent architectures with Denaria: the recommendation that institutions must be redesigned for verification, accountability, interpretability, and dual-use safety rests on the assumption that such systems can be engineered to deliver these properties reliably, yet the manuscript provides only high-level illustrative sketches without addressing implementation challenges or failure modes for these properties.

    Authors: We agree that the Denaria sketch is high-level and illustrative rather than a full engineering specification, and that the manuscript does not provide detailed implementation pathways or exhaustive failure-mode analysis. As a position paper, its primary aim is to argue that the transition to AI scientists is qualitative and that institutional redesign must occur in parallel with technical development; the sketches serve to make the conceptual architecture concrete enough to ground the normative claims. The recommendation for reformed institutions is motivated precisely by the recognition that reliable engineering of verification, accountability, interpretability, and dual-use safety is non-trivial and cannot be assumed under current structures. That said, the comment correctly identifies a gap: the text would benefit from an explicit acknowledgment of key implementation challenges (e.g., ensuring verifiable provenance in multi-agent loops, handling emergent misalignment, and scaling interpretability techniques). We will therefore make a partial revision by adding a short subsection in the Denaria discussion that flags representative challenges and failure modes without attempting to solve them, thereby clarifying the illustrative nature of the sketch and the need for subsequent engineering work. revision: partial

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript is a conceptual position paper with no equations, fitted parameters, formal derivations, or quantitative predictions. Its central claims are normative arguments about institutional redesign and the qualitative nature of AI-assisted discovery; these do not reduce to self-defined inputs, fitted data, or self-citation chains. The illustrative reference to Denaria is presented as a sketch rather than a load-bearing derivation. No steps match any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The argument rests on domain assumptions about AI capabilities and the feasibility of institutional redesign without independent evidence supplied in the abstract.

axioms (1)
  • domain assumption Agentic AI systems can perform tasks spanning literature synthesis, code generation, data analysis, hypothesis proposal, and model criticism at a level that supports qualitative expansion of discovery capacity.
    Invoked as the premise for treating the transition as qualitative.

pith-pipeline@v0.9.1-grok · 5720 in / 1073 out tokens · 30104 ms · 2026-06-26T08:43:09.185860+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 5 canonical work pages

  1. [1]

    The indiscriminate adoption of ai threatens the foundations of academia

    Roberto Trotta. The indiscriminate adoption of ai threatens the foundations of academia. Nature Astronomy, 9 0 (12): 0 1748--1749, December 2025. doi:10.1038/s41550-025-02738-w

    work page doi:10.1038/s41550-025-02738-w 2025

  2. [2]

    Scientific production in the era of large language models

    Keigo Kusumegi, Xinyu Yang, Paul Ginsparg, Mathijs de Vaan, Toby Stuart, and Yian Yin. Scientific production in the era of large language models. arXiv preprint arXiv:2601.13187, 2026. URL https://arxiv.org/abs/2601.13187

    arXiv 2026

  3. [3]

    LLM hallucinations in the wild: Large-scale evidence from non-existent citations

    Zhenyue Zhao, Yihe Wang, Toby Stuart, Mathijs De Vaan, Paul Ginsparg, and Yian Yin. LLM hallucinations in the wild: Large-scale evidence from non-existent citations . 2026

    2026

  4. [4]

    Hiranya V. Peiris. Large language models are not the problem. Nature Astronomy, April 2026. doi:10.1038/s41550-026-02837-2

    work page doi:10.1038/s41550-026-02837-2 2026

  5. [5]

    Ai can help scientists publish less

    Gianfranco Bertone. Ai can help scientists publish less. Nature Astronomy, 10: 0 557--559, 2026. doi:10.1038/s41550-026-02720-0. URL https://nature.com

    work page doi:10.1038/s41550-026-02720-0 2026

  6. [6]

    Alireza Ghafarollahi and Markus J. Buehler. Sciagents: Automating scientific discovery through multi-agent intelligent graph reasoning. 2024. URL https://arxiv.org/abs/2409.05556

    arXiv 2024

  7. [7]

    The ai scientist: Towards fully automated open-ended scientific discovery

    Chris Lu, Cheng Lu, Robert Tjarko Lange, Jakob Foerster, Jeff Clune, and David Ha. The ai scientist: Towards fully automated open-ended scientific discovery. 2024. URL https://arxiv.org/abs/2408.06292

    Pith/arXiv arXiv 2024

  8. [8]

    The ai scientist-v2: Workshop-level automated scientific discovery via agentic tree search

    Yutaro Yamada, Robert Tjarko Lange, Chris Lu, Shengran Hu, Cheng Lu, Jakob Foerster, et al. The ai scientist-v2: Workshop-level automated scientific discovery via agentic tree search. 2025. URL https://arxiv.org/abs/2504.08066

    Pith/arXiv arXiv 2025

  9. [9]

    Towards an ai co-scientist

    Juraj Gottweis, Wei-Hung Weng, Alexander Daryin, Tao Tu, Anil Palepu, Petar Sirkovic, Artiom Myaskovsky, Felix Weissenberger, Keran Rong, Ryutaro Tanno, Khaled Saab, Dan Popovici, Jacob Blum, Fan Zhang, Katherine Chou, Avinatan Hassidim, Burak Gokturk, Amin Vahdat, Pushmeet Kohli, et al. Towards an ai co-scientist. 2025. URL https://arxiv.org/abs/2502.18864

    Pith/arXiv arXiv 2025

  10. [10]

    Bayer, Aidan Acquah, Chetana Amancharla, Almog Barzilay-Siegal, Pablo Bermejo, Camille Bilodeau, Pablo Cardenas Ramirez, Miles Cranmer, Urbano L

    Francisco Villaescusa-Navarro, Boris Bolliet, Pablo Villanueva-Domingo, Adrian E. Bayer, Aidan Acquah, Chetana Amancharla, Almog Barzilay-Siegal, Pablo Bermejo, Camille Bilodeau, Pablo Cardenas Ramirez, Miles Cranmer, Urbano L. Franca, ChangHoon Hahn, Yan-Fei Jiang, Raul Jimenez, Jun-Young Lee, Antonio Lerario, Osman Mamun, Thomas Meier, Anupam A. Ojha, P...

    arXiv 2025

  11. [11]

    Ai--assisted exploration: Dhost theories without quantum ghosts

    Ginevra Braga, Raul Jimenez, and Sabino Matarrese. Ai--assisted exploration: Dhost theories without quantum ghosts. 2026. URL https://arxiv.org/abs/2604.16531

    Pith/arXiv arXiv 2026

  12. [12]

    Karl R. Popper. Conjectures and Refutations: The Growth of Scientific Knowledge. Routledge, London, 1963. ISBN 0-415-04318-2

    1963

  13. [13]

    Kauffman

    Stuart A. Kauffman. Investigations. Oxford University Press, Oxford and New York, 2000. ISBN 9780195121049

    2000

  14. [14]

    Sygdommen til D den

    S ren Kierkegaard. Sygdommen til D den . C.A. Reitzel, K benhavn, 1849. Published under the pseudonym Anti-Climacus

  15. [15]

    Fyodor Dostoevsky. Idiot. Russkiy Vestnik (The Russian Messenger), 1868--1869. Originally published in serial form in Russian

  16. [16]

    Sources of the Self: The Making of the Modern Identity

    Charles Taylor. Sources of the Self: The Making of the Modern Identity. Harvard University Press, Cambridge, MA, 1989

    1989

  17. [17]

    Ai, human cognition and knowledge collapse

    Daron Acemoglu et al. Ai, human cognition and knowledge collapse. Working Paper 34910, National Bureau of Economic Research, 2026. URL https://www.nber.org/papers/w34910

    2026

  18. [18]

    Against Method: Outline of an Anarchistic Theory of Knowledge

    Paul Feyerabend. Against Method: Outline of an Anarchistic Theory of Knowledge. Verso, London, 1975

    1975

  19. [19]

    Computer Power and Human Reason: From Judgment to Calculation

    Joseph Weizenbaum. Computer Power and Human Reason: From Judgment to Calculation. W. H. Freeman, San Francisco, 1976

    1976

  20. [20]

    Researchers who use hallucinated references to face arXiv ban

    Dalmeet Singh Chawla. Researchers who use hallucinated references to face arXiv ban . Nature, May 2026. doi:10.1038/d41586-026-01595-5. URL https://www.nature.com/articles/d41586-026-01595-5

    work page doi:10.1038/d41586-026-01595-5 2026

  21. [21]

    Superintelligence: Paths, dangers, strategies

    Nick Bostrom. Superintelligence: Paths, dangers, strategies. Oxford University Press, Oxford, UK, 2014. ISBN 978-0-19-967811-2

    2014

  22. [22]

    Alan M. Turing. Intelligent machinery, a heretical theory. In B. Jack Copeland, editor, The Essential Turing: Seminal Writings in Computing, Logic, Philosophy, Artificial Intelligence, and Artificial Life, pages 465--475. Oxford University Press, Oxford, 2004. doi:10.1093/oso/9780198250791.003.0018. Originally written c. 1951

    work page doi:10.1093/oso/9780198250791.003.0018 2004

  23. [23]

    Allons-nous continuer la recherche scientifique ? Conference au CERN, Geneve, 1972

    Alexandre Grothendieck. Allons-nous continuer la recherche scientifique ? Conference au CERN, Geneve, 1972. URL https://webusers.imj-prg.fr/ leila.schneps/grothendieckcircle/Allonsnous.pdf. Transcription de la conference du 27 janvier 1972

    1972

  24. [24]

    Reviewertoo: Should ai join the program committee? a look at the future of peer review, 2025

    Gaurav Sahu, Hugo Larochelle, Laurent Charlin, and Christopher Pal. Reviewertoo: Should ai join the program committee? a look at the future of peer review, 2025. URL https://arxiv.org/abs/2510.08867

    arXiv 2025

  25. [25]

    Your brain on chatgpt: Accumulation of cognitive debt when using an ai assistant for essay writing task, 2025

    Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashly Vivian Beresnitzky, Iris Braunstein, and Pattie Maes. Your brain on chatgpt: Accumulation of cognitive debt when using an ai assistant for essay writing task, 2025. URL https://arxiv.org/abs/2506.08872

    Pith/arXiv arXiv 2025

  26. [26]

    Natural emergent misalignment from reward hacking in production rl, 2025

    Chris MacDiarmid et al. Natural emergent misalignment from reward hacking in production rl, 2025

    2025

  27. [27]

    Thomas S. Kuhn. The Structure of Scientific Revolutions. University of Chicago Press, Chicago, IL, 1962

    1962

  28. [28]

    Republic

    Plato. Republic. -380. Ancient Greek philosophical text, traditionally dated to c. 380 BCE